The use of WER (Word Error Rate) in the B2B sector

Automatic Speech Recognition (ASR) has transformed the way we interact with technology, making our lives easier in various aspects such as customer service, meeting transcription, and the automation of administrative tasks. However, the accuracy of ASR transcriptions still presents challenges, as its error rate is three times higher than that of human transcribers.

We will delve into how the use of WER in the B2B sector allows companies to evaluate and compare the effectiveness of different ASR systems based on their specific needs.

What is WER?

The WER (Word Error Rate) is a metric used in the field of Automatic Speech Recognition (ASR) to measure the accuracy of voice to text conversion. This metric assesses the number of transcription errors in relation to the total number of spoken words, considering substitutions, insertions, and deletions.

To calculate the WER, the total number of errors is divided by the total number of spoken words. A low WER generally indicates higher ASR software accuracy, while a high WER suggests lower accuracy.

How WER is calculated

Here is a simple formula to understand how the Word Error Rate (WER) is calculated:


  1. S stands for substitutions,
  2. I stands for insertions,
  3. D stands for deletions,
  4. N is the total number of spoken words

This measure is based on the Levenshtein distance, which measures the difference between two strings of words in a transcription. For example, if a transcription has 9 errors in a 36-word phone call, the WER would be 25%.

How can WER help decrease inaccuracies in B2B communications?

In the B2B sector, precise and efficient communication is crucial for business success. By using WER as a performance indicator, companies can objectively analyze and compare the accuracy of different ASR systems. This analysis allows identifying which solutions best fit their requirements, resulting in higher efficiency and quality in voice-to-text transcription.

By comparing the WER between different ASR systems, companies can identify areas for improvement in their current solutions and seek systems that address these shortcomings. However, it’s crucial to consider the limitations of WER when evaluating and comparing ASR systems in the B2B sector, as it doesn’t take into account the source of the errors nor the importance of the words in the transcription.

Companies should complement this metric with a qualitative analysis that includes factors such as audio quality, background noise, and industry-specific vocabulary. To reduce errors, actions such as improving audio quality, training the model with more data, limiting vocabulary, and applying grammar and context, which will allow better understanding of communication and reduction of recognition errors.

Therefore, it is essential to select a tool that suits the specific needs of your business and the audio data that will be analyzed. Upbe‘s ASR is designed to transcribe telephone dialogues. It is specifically trained for a sales context where there may be background noise, voice overlap, and limited quality recordings.

Do you have questions? Contact us and we will answer them!

Leave a comment